PhD student in Computer Science, University of Cambridge

Mantas Bakšys

I am a PhD student at the Computer Lab, University of Cambridge supervised by Prof. Sean Holden. My research interests lie in the broad areas of Automated Theorem Proving and Formal Mathematics. In particular, I am interested in application of Machine Learning for proof-discovery in the setting of interactive theorem provers. I have experience working with reinforcement learning techniques for formal reasoning, having contributed to state-of-the-art models for autoformalization and proving in Lean. Feel free to reach out if you find my research interesting!

Interests
  • Automated theorem proving using Reinforcement Learning
  • Formal theorem proving and verification
  • Interactive theorem provers and their ecosystems
Education
  • PhD, Computer Science, 2024-2028 (expected)

    University of Cambridge

  • MMath, Part III Mathematics, 2023-2024

    University of Cambridge

  • BA (Hons), Mathematics, 2020-2023

    University of Cambridge

Publications

(2025). Kimina Lean Server: Technical Report. arXiv preprint arXiv:2504.21230.

PDF Cite arXiv

(2025). Kimina-Prover Preview: Towards Large Formal Reasoning Models with Reinforcement Learning. arXiv preprint arXiv:2504.11354.

PDF Cite arXiv

(2025). Kimina-Prover: Applying Test-time RL Search on Large Formal Reasoning Models. HuggingFace Blog.

Cite HuggingFace Blog

(2023). A Formalisation of the Balog–Szemerédi–Gowers Theorem in Isabelle/HOL. Proceedings of the 12th ACM SIGPLAN International Conference on Certified Programs and Proofs.

Cite DOI URL

(2023). Formal Mathematics Statement Curriculum Learning. Top 25% paper at ICLR 2023.

Cite URL

Research Experience

 
 
 
 
 
Applied Scientist Intern
July 2025 – Present London, UK
  • Supervised by Dr. Stefan Zetsche, working on Reinforcement Learning for program verification
  • Developing novel RL approaches for automated theorem proving and formal verification
 
 
 
 
 
Researcher
September 2024 – Present Remote
  • Using RL techniques to train AI models for Interactive Theorem Proving in Lean
  • Development of Lean-Python interfaces for dataset and benchmark creation and curation
  • Released open-source models for autoformalization and proving in Lean with state-of-the-art performance on the miniF2F benchmark
 
 
 
 
 
Student Researcher
July 2023 – October 2023 Cambridge, UK
  • Worked in a research group led by Professor Timothy Gowers on automated theorem-proving
  • Focus on motivated proof discovery using the Lean theorem prover
  • Jointly developed an interactive Lean program to guide further research in human-oriented theorem proving
 
 
 
 
 
Student Researcher
June 2022 – August 2022 Cambridge, UK
  • Worked in the ALEXANDRIA group supervised by Dr Angeliki Koutsoukou-Argyraki
  • Successful formalisation of Master’s level material in Additive Combinatorics including the Balog-Szemeredi-Gowers theorem
  • Co-authored a paper accepted to CPP 2023
 
 
 
 
 
Research Intern
December 2021 – January 2022 Remote
  • Worked with Stanislas Polu on training ML models to find formal proofs using the Lean theorem prover
  • Analyzed properties of trained models and investigated new research directions
  • Co-authored a paper accepted to ICLR 2023
 
 
 
 
 
Student Researcher
June 2021 – August 2021 Cambridge, UK
  • Worked with Dr Aled Walker on a Summer Research Placement spanning 8 weeks
  • Investigated the Multiplication Table problem for bipartite graphs
  • Co-authored a paper pre-print

Talks

Invited Talk on Kimina Prover

Joint invited talk with Jonas Bayer presenting Kimina Prover. AITP 2025 Conference, Aussois, France, September 2025.

Kimina Prover: Test-time RL enabled Theorem Proving with Large Formal Reasoning Models

Presented Kimina Prover to the AWS Reasoning Group. Amazon Web Services (AWS), London, UK, August 2025.